Perbandingan Metode Autoregressive Integrated Moving Average Exogenous (ARIMAX), Long Short-Term Memory (LSTM), dan Extreme Gradient Boosting (XGBoost) dalam Peramalan Data Runtun Waktu (Studi Kasus: Data Seismik Gempa Bumi di Indonesia Tahun 2004 - 2024)

Najzwa Hanifah Azkarrula, Prof. Dr. Drs. Gunardi, M.Si.

2025 | Skripsi | STATISTIKA

Gempa bumi merupakan fenomena alam yang sering terjadi di Indonesia, mengingat letaknya yang berada di Cincin Api Pasifik. Potensi gempa megathrust turut menjadi perhatian utama, terutama sejak bencana Aceh 2004 dan peringatan BMKG pada Agustus 2024 terkait seismic gap. Penelitian ini bertujuan untuk membandingkan pendekatan metode pembelajaran statistik Autoregressive Integrated Moving Average Exogenous (ARIMAX), metode pembelajaran mendalam Long Short-Term Memory (LSTM), dan metode ansambel Extreme Gradient Boosting (XGBoost) dalam meramalkan magnitudo gempa bumi menggunakan data seismik Indonesia periode 2004–2024 dari United States Geological Survey (USGS). Data yang digunakan meliputi fitur latitude, longitude, dan depth untuk memprediksi magnitudo. Ketiga metode diimplementasikan dengan pendekatan runtun waktu: ARIMAX berbasis statistik, LSTM berbasis jaringan saraf, dan XGBoost berbasis gradient boosting. Hasil analisis menunjukkan XGBoost memiliki akurasi tertinggi dengan Root Mean Square Error (RMSE) 0.0667, Mean Squared Error (MSE) 0.0045, Mean Absolute Error (MAE) 0.0506, dan Mean Absolute Percentage Error (MAPE) 1.1285%, diikuti LSTM (RMSE 0.0709) dan ARIMAX (RMSE 0.0745). Prediksi magnitudo untuk sepuluh minggu ke depan (September–November 2024) menggunakan XGBoost menghasilkan nilai stabil antara 4.44–4.55, konsisten dengan pola seismik Indonesia. Penelitian ini menyimpulkan bahwa XGBoost adalah metode paling akurat untuk peramalan magnitudo gempa, memberikan kontribusi signifikan bagi mitigasi bencana, kesiapsiagaan masyarakat, dan perencanaan infrastruktur tahan gempa di Indonesia.

Earthquakes are a frequent natural phenomenon in Indonesia, given its location within the Pacific Ring of Fire. The potential for megathrust earthquakes has garnered significant attention, particularly since the 2004 Aceh disaster and the BMKG’s August 2024 warning regarding the seismic gap. This study aims to compare the statistical learning approach of Autoregressive Integrated Moving Average Exogenous (ARIMAX), the deep learning method Long Short-Term Memory (LSTM), and the ensemble method Extreme Gradient Boosting (XGBoost) in forecasting earthquake magnitude using seismic data from Indonesia spanning 2004–2024, sourced from the United States Geological Survey (USGS). The data includes features such as latitude, longitude, and depth to predict magnitude. These methods were implemented using a time-series approach: ARIMAX based on statistics, LSTM based on neural networks, and XGBoost based on gradient boosting. Results indicate that XGBoost achieved the highest accuracy with a Root Mean Square Error (RMSE) of 0.0667, Mean Squared Error (MSE) of 0.0045, Mean Absolute Error (MAE) of 0.0506, and Mean Absolute Percentage Error (MAPE) of 1.1285%, followed by LSTM (RMSE 0.0709) and ARIMAX (RMSE 0.0745). Magnitude forecasts for the next ten weeks (September–November 2024) using XGBoost yielded stable values ranging from 4.44 to 4.55, consistent with Indonesia’s seismic patterns. This study concludes that XGBoost is the most accurate method for forecasting earthquake magnitude, significantly contributing to disaster mitigation, community preparedness, and the planning of earthquakeresistant infrastructure in Indonesia.

Kata Kunci : Gempa Bumi, Magnitudo, Peramalan, ARIMAX, Long Short-Term Memory (LSTM), Extreme Gradient Boosting (XGBoost), Earthquake, Magnitude, Forecasting